A Q-Learning-based Decision Making Scheme for Application Reconfiguration in Sensor Networks

Application reconfiguration is a principal method of improving the application flexibility and adaptability to dynamic environment in sensor networks. Motivated by this, we propose a novel Q-learning-based decision making scheme in sensor networks to support the environment adaptive reconfiguration decision. In this scheme, we introduce the rule-based reasoning to predict the trend of environmental change and make reconfiguration decision adoptively. In particular, we apply a real-time reinforcement learning algorithm, called Q-learning, to evaluate the certainty value of the rules in this decision mechanism. Moreover, we map the continuous state space of environment condition into a discrete and finite state space to design a certainty function. The simulation results show that the Q-learning-based decision mechanism can not only learn the change of environment, but also effectively make the reconfiguration decision in sensor networks.